byte of 87: FlameGaurd

• 

  FlameGaurd Prototype

Inspiration

In 2021 alone, Statistics Canada reported 15,395 indoor fires, highlighting the urgent need for better fire detection and response systems. Fire incidents can escalate quickly if the wrong suppression method is used. Not all fires are the same—water can worsen grease fires, and some electrical fires need specific extinguishers.

However, studies suggest that over 60% of individuals do not know which extinguisher to use in an emergency, increasing the risk of serious injury and fire spreading. We wanted to create a system that not only detects fire but also classifies it and provides immediate, accurate response instructions to prevent further damage and save lives.

What it does

FlameGuard uses computer vision to detect and classify different types of fires in real-time. Based on the fire type, it provides recommendations on the safest extinguishing method—whether it's water (Class A), foam (Class A & B), CO₂ (Class B & C), dry chemical (Class A, B, C), or specialized agents (Class D & K). This ensures that responders and individuals make informed decisions when dealing with a fire emergency, reducing the risk of worsening the situation.

How we built it

Computer Vision: We used OpenCV and YOLOv8 to train a model that can distinguish between different types of fires. Dataset: We sourced and preprocessed fire-related images to train our model for accurate classification. Real-Time Processing: Using a Raspberry Pi camera and servo motor to continuously scans for fire outbreaks. Recommendation System: Once a fire is detected, FlameGuard identifies its class and provides real-time safety instructions on the appropriate extinguishing method. The guidance is delivered via a speaker, ensuring users can take immediate and informed action.

Challenges we ran into

Dataset Availability: Finding diverse, high-quality fire images for training was challenging. We had to preprocess and augment data for better model performance. Real-Time Processing: Ensuring low-latency classification while maintaining accuracy required optimization. False Positives: Differentiating between harmless warm light sources (like candles) and actual fires was tricky. We refined our model to improve precision.

Accomplishments that we're proud of

Successfully trained a model to classify fire types with high accuracy. Implemented a real-time detection system that provides actionable safety recommendations. Optimized our model to minimize false positives and improve response time.

What we learned

The importance of dataset quality in training a robust AI model. Techniques to improve real-time performance using model optimization and lightweight architectures. The significance of fire safety knowledge and how AI can assist in emergency situations.

What's next for FlameGuard

Drone/UAV integration: We aim to integrate FlameGuard with drones to scan large areas and detect early-stage forest fires before they spread uncontrollably. Our system can help authorities respond faster and minimize damage. Mobile & Web App: Expanding accessibility so users can receive fire alerts and safety guidance on their devices. Advanced Fire Classification: Enhancing the model to detect smoke and fire spread patterns for better predictions. Integration with Emergency Services: Sending real-time alerts to firefighters with fire type classification to aid rapid response.

FlameGuard is more than just fire detection—it's about making fire safety smarter and more accessible.

Built With

• opencv
• python
• raspberry-pi
• yolov8